# Dr Yi Ma

### Biography

Dr. Yi Ma is a Reader within the Institute for Communication Systems (ICS, formerly the Centre for Communication Systems Research, CCSR). He has extensive expertise in the areas of Signal Processing, Machine Learning and Information Theory, with their applications in Telecommunications.

### Research

### Research interests

- Machine learning for future physical layer design
- Transceiver optimization for future communication systems such as URLLC.
- Scalable Distributed MIMO technology
- Opportunistic networking and cooperative communications
- Hybrid data fusion and machine learning for mobile localization
- Estimation, detection and synchronization
- Information theory and coding

**Research projects**

- 2017-2021 Artificial intelligence for future communications (Industry fund)
- 2013-2016 RESCUE (FP7 ICT Consortium)
- 2013-2014 LTE Machine-Type Communications: Phase II (Industry fund)
- 2012-2013 LTE Machine-Type Communications: Phase I (Industry fund)
- 2010-2013 WHERE2 (FP7 ICT Consortium)
- 2011-2012 Wi-Fi Indoor Positioning (EPSRC)
- 2010-2013 EXALTED (FP7 ICT Consortium)
- 2007-2010 WHERE (FP7 ICT Consortium)
- 2005-2007 WINNER2 (FP6 ICT Consortium)
- 2004-2005 4MORE (FP6 ICT Consortium)
- 2006-2009 Mobile VCE-Core 4 (EPSRC)

**PhD Position**

I am constantly looking for well self-motivated PhD candidates with excellent background in Physics, Mathematics, Wireless Communications, or Computer Science. Prior to submit your application, please make sure you have met the following University requirements :

- A 1st class BSc degree or MSc with a Distinction (or equivalent to top 10% internationally).
- A good research proposal (if ICS funding support is requested, please clearly indicate why the proposed research should be financially supported by the ICS).
- For international students, it is essential to meet the University's English requirements (IELTS 6.5 or above (overall) with each section of 6.0 or above).

### Supervision

## Postgraduate research supervision

**Current PhD Students:**

- Songyan Xue: Deep learning for future modem design
- Ang Li: Deep learning for NOMA
- Lifu Liu: Low-cost mmWave solutions
- Jinfei Wang: Ultra-reliable low-latency communications (URLLC)

**Ex-PhD Students:**

- Hongju Liu (04-08): Channel estimation for multicarrier transmissions
- Na Yi (06-09): Cooperative communications
- Yuanyuan Zhang (06-10): Adaptive cooperative relays
- Mohammad Movahhedian (07-10): Frequency synchronization for multiuser multicarrier transmissions.
- Parisa Cheraghi (09-12): Advanced spectrum sensing techniques
- Ziming He (08-12): Advanced mobile positioning and tracking techniques
- Hui Luo (07-11): Cooperative communications for satellite systems
- Zhengwei Lu (09-13): Pilot-assisted fast spectrum sensing techniques
- Jiancao Hou (10-14): Advanced multiuser-MIMO transmitter design
- Chuyi Qian (10-14): Opportunistic relaying protocols
- Erik Yngvesson (13-17): Coexistence of Massive MIMO in unlicensed bands
- Juan Carlos De Luna Ducoing (14-17): Advanced modulations for scalable multiuser MIMO
- Abdullah Alonazi (11-16): Less-calibrated indoor mobile localization
- Guangyi Wang (12-17): Estimation of pilot contaminated channels
- Raouf Yamani (15-17): Low-complexity vector perturbation for MIMO nonlinear precoding

### My teaching

- EEEM017: Fundamentals of Mobile Communications
- EEE3006: Digital Communications
- Personal and tutorial tutor for undergraduate students.
- Year 1 and Year 2 undergraduate examination officer

### My publications

### Publications

scheduling approach to enhance the random beamforming (RBF)

with limited feedback in multiple-input ?multiple-output (MIMO)

broadcast channels. Such scheme was shown to obtain the optimal

scaling law of sum rate in the large number of user regime.

However, for a small number of user cases, the system degrees-of freedom cannot be well exploited, and the accuracy of predicting

users? signal-to-interference-plus-noise ratios (SINRs) is also degraded.

Motivated by these, two strategies are proposed to solve

the problems. Specifically, the conventional spatial domain RBF is

first extended to the space?frequency domain. The first strategy

aims to maximize the number of active users based on users?

initially predicted SINRs; the second strategy is to schedule the

maximum number of users, whose preferred beams can coexist

with each others, with more accurate users? SINRs prediction

method. Computer simulations are carried out to examine the proposed

strategies in terms of the sum rate and the number of active

users. It is shown that the first strategy achieves close performance

to the corresponding brute-force search with lower complexity.

Moreover, the second strategy improves the performance by accurately

predicting users? SINRs at the price of relatively increased

complexity and feedback overhead.

The question of how these properties can be harnessed is explored by considering two perspectives: no cooperation and cooperation between users. For the cooperative scenario, a spatial-domain interweave spectrum sharing scheme is introduced that enables opportunistic transmission at a controlled cost to the license holders. The proposed scheme demonstrates three excellent characteristics: that exploitation of the spatial domain allows opportunistic communication in a ?spatial hole,? that spectrum sharing is effectively enabled by inter-tier cooperation, and finally that in this scenario spatial-domain interweave is feasible with a ?small? (as compared to the number of receive antennas at the incumbent) number of transmit antennas. In essence, this opens the possibility of the incumbents? performance to be traded against opportunistic transmission. In the non-cooperative scenario, a spectrum sharing model between a small and large MU-MIMO system is proposed and analysed. The significant service antenna number asymmetry poses unique challenges and opportunities. In the limit of an infinite number of service antennas at one of the access point, the interference and noise power tends to zero and the transmit power can also be scaled back accordingly. These traits seem ideal for use in a spectrum sharing scenario, but in the present case with the coexistence of a conventional MIMO system and with a finite number of service antennas, how will the system behave? The resulting interference scenario is analysed explicitly both in the uplink and downlink, assuming linear receive and transmit equalizers, respectively. Characterization of the mean SINR operating point and required transmit power are presented, and concise transmit power scaling laws are derived. The scaling laws offer insight into how the system behaves with the number of service antennas and system load.

First, a technique that uses real-valued modulation in fully- and over-loaded cases in large MU-MIMO systems, where there are equal or more UTs than service antennas. It is seen that the use of real constellations with a widely linear equaliser benefits from an increased spatial diversity gain over complex constellations with a linear equaliser. Moreover, a likelihood ascent search (LAS) algorithm post-processing stage is applied to further improve the error performance. Computer simulations show remarkable results for large MU-MIMO sizes in uncoded or coded cases.

Second, recognising that real-valued modulation offers poor modulation efficiency, a real-complex hybrid modulation (RCHM) scheme is proposed, where a mix of real- and complex-valued symbols are interleaved in the spatial and temporal domains. It is seen that RCHM combines the merits of real and complex modulations and enables the adjustment of the diversity-multiplexing tradeoff. Through the system outage probability analysis, the optimal ratio of the number real-to-complex symbols, as well as their optimal power allocation, is found for the RCHM pattern. Furthermore, reliability is improved with a small expense in complexity through the use of a successive interference cancellation (SIC) stage. Results are validated through the mathematical analysis of the average bit error rate and through computer simulations considering single and multiple base station scenarios, which show SNR gains over conventional approaches in excess of 5 dB at 1% BLER.

The results suggest that an expense in complexity is not the only way to improve error performance, but near-optimal reliability is also possible using simple techniques through a reduction in the multiplexing gain. Therefore, rather than a two-way complexity vs. performance tradeoff in MU-MIMO detection, a three-way tradeoff may be more appropriate, and is roughly expressed in the following statement:

?Low complexity, high reliability, high multiplexing gain: choose two.?

hybrid modulation (RCHM), is proposed to scale up

multiuser multiple-input multiple-output (MU-MIMO) detection

with particular concern on the use of equal or approximately

equal service antennas and user terminals. By RCHM, we mean

that user terminals transmit their data sequences with a mix of

real and complex modulation symbols interleaved in the spatial

and temporal domain. It is shown, through the system outage

probability, RCHM can combine the merits of real and complex

modulations to achieve the best spatial diversity-multiplexing

trade-off that minimizes the required transmit-power given a

sum-rate. The signal pattern of RCHM is optimized with respect

to the real-to-complex symbol ratio as well as power allocation.

It is also shown that RCHM equips the successive interference

canceling MU-MIMO receiver with near-optimal performances

and fast convergence in Rayleigh fading channels. This result is

validated through our mathematical analysis of the average biterror-

rate as well as extensive computer simulations considering

the case with single or multiple base-stations.

estimation method for generalized MC-CDMA systems in unknown frequency-selective channels utilizing hidden pi-

lots. It is established that CFO is identifiable in the frequency domain by employing cyclic statistics (CS) and linear re-gression (LR) algorithms. We show that the CS-based estimator is capable of mitigating the normalized CFO (NCFO) to a small error value. Then, the LR-based estimator can be employed to offer more accurate estimation by removing the residual quantization error after the CS-based estimator.

for full-duplex (FD) relaying networks is proposed to mitigate

error propagation effects and improve system spectral efficiency.

The idea is to allow the FD relay node to predict the correctly

decoded symbols of each frame, based on the generalized square

deviation method, and discard the erroneously decoded symbols,

resulting in fewer errors being forwarded to the destination node.

Using the capability for simultaneous transmission and reception

at the FD relay node, our proposed strategy can improve the

transmission efficiency without extra cost of signalling overhead.

In addition, targeting on the derived expression for outage probability,

we compare it with half-duplex (HD) relaying case, and

provide the transmission power and relay location optimization

strategy to further enhance system performances. The results

show that our proposed scheme outperforms the classic relaying

protocols, such as cyclic redundancy check based selective

decode-and-forward (S-DF) relaying and threshold based SDF

relaying in terms of outage probability and bit-error-rate.

Moreover, the performances with optimal power allocation are

better than those with equal power allocation, especially when

the FD relay node encounters strong self-interference and/or it

is close to the destination node.

synchronization problem inherent in orthogonal

frequency-division multiple access (OFDMA) uplink

communications, where the carrier frequency offset (CFO)

for each user may be different, and they can be hardly

compensated at the receiver side. Our major contribution

lies in the development of a novel OFDM receiver that

is resilient to unknown random CFO thanks to the use

of a CFO-compensator bank. Specifically, the whole CFO

range is evenly divided into a set of sub-ranges, with

each being supported by a dedicated CFO compensator.

Given the optimization for CFO compensator a NP-hard

problem, a machine deep-learning approach is proposed

to yield a good sub-optimal solution. It is shown that the

proposed receiver is able to offer inter-carrier interference

free performance for OFDMA systems operating at a wide

range of SNRs.

and noncoherent receiver optimization for multiuser single-input

multiple-output (MU-SIMO) communications through unsupervised

deep learning. It is shown that MU-SIMO can be modeled

as a deep neural network with three essential layers, which

include a partially-connected linear layer for joint multiuser

waveform design at the transmitter side, and two nonlinear layers

for the noncoherent signal detection. The proposed approach

demonstrates remarkable MU-SIMO noncoherent communication

performance in Rayleigh fading channels.

a mobile device, with a single RF chain, shares its message

with a set of mobile devices through narrowband mmWave

channel, an analogue-beam splitting approach is proposed

to achieve a good capacity and coverage trade-off. The

proposed approach aims at maximizing the capacity of

the mmWave multicast channel through antenna-element

grouping and adaptive phase shifting, which takes into

account of the inter-beam interference. When receivers are

randomly distributed on a circle centered at the transmitter,

according to the uniform distribution, it is found

that the impact of inter-beam interference on the channel

capacity can be negligibly small, and thus the analoguebeam

splitting approach can be largely simplified in practice.

Computer simulations are carried out to elaborate our

theoretical study and demonstrate considerable advantages

of the proposed analogue-beam splitting approach.

approach is proposed to tackle the multiuser frequency synchronization

problem inherent in orthogonal frequency-division

multiple-access (OFDMA) uplink communications. The key idea

lies in the use of the feed-forward deep neural network (FF-DNN)

for multiuser interference (MUI) cancellation taking advantage

of their strong classification capability. Basically, the proposed

FF-DNN consists of two essential functional layers. One is

called carrier-frequency-offsets (CFOs) classification layer that

is responsible for identifying the users? CFO range, and another

is called MUI-cancellation layer responsible for joint multiuser

detection (MUD) and frequency synchronization. By such means,

the proposed FF-DNN approach showcases remarkable MUIcancellation

performances without the need of multiuser CFO

estimation. In addition, we also exhibit an interesting phenomenon

occurred at the CFO-classification stage, where the

CFO-classification performance get improved exponentially with

the increase of the number of users. This is called multiuser

diversity gain in the CFO-classification stage, which is carefully

studied in this paper.

for multiuser single-input multiple-output (MU-SIMO) coherent

detection are extensively investigated. According to the ways

of utilizing the channel state information at the receiver side

(CSIR), deep learning solutions are divided into two groups.

One group is called equalization and learning, which utilizes the

CSIR for channel equalization and then employ deep learning for

multiuser detection (MUD). The other is called direct learning,

which directly feeds the CSIR, together with the received signal,

into deep neural networks (DNN) to conduct the MUD. It is found

that the direct learning solutions outperform the equalizationand-

learning solutions due to their better exploitation of the

sequence detection gain. On the other hand, the direct learning

solutions are not scalable to the size of SIMO networks, as

current DNN architectures cannot efficiently handle many cochannel

interferences. Motivated by this observation, we propose

a novel direct learning approach, which can combine the merits

of feedforward DNN and parallel interference cancellation. It is

shown that the proposed approach trades off the complexity for

the learning scalability, and the complexity can be managed due

to the parallel network architecture.

that is employed to reconstruct original information bits from

a non-recursive convolutional codeword in noise, with the goal

of reducing the decoding latency without compromising the

performance. This goal is achieved by means of cutting a

received codeword into a number of sub-codewords (SCWs)

and feeding them into a two-stage decoder. At the first stage,

SCWs are decoded in parallel using the Viterbi algorithm or

equivalently the brute force algorithm. Major challenge arises

when determining the initial state of the trellis diagram for each

SCW, which is uncertain except for the first one; and such results

in multiple decoding outcomes for every SCW. To eliminate or

more precisely exploit the uncertainty, an Euclidean-distance

minimization algorithm is employed to merge neighboring SCWs;

and this is called the merging stage, which can also run in

parallel. Our work reveals that the proposed two-stage decoder

is optimal and has its latency growing logarithmically, instead

of linearly as for the Viterbi algorithm, with respect to the

codeword length. Moreover, it is shown that the decoding latency

can be further reduced by employing artificial neural networks

for the SCW decoding. Computer simulations are conducted

for two typical convolutional codes, and the results confirm our

theoretical analysis.

particularly on the physical layer design; and what penalties there may have? These questions motivate a fundamental rethinking of the wireless modem design in the artificial intelligence era. Through several physical-layer case studies, we argue for a significant role that machine learning could play, for instance in parallel error-control coding and decoding, channel equalization, interference cancellation,

as well as multiuser and multiantenna detection. In addition, we will also discuss the fundamental bottlenecks of machine learning as

well as their potential solutions in this paper.

Method for Dynamic Environments,IEEE ICC 2020

aerial vehicles (UAVs) as flying base stations (BSs) for

optimizing the throughput of mobile users is investigated for UAV networks. This problem is formulated as a time-varying mixed-integer non-convex programming (MINP) problem, which is challenging to find an optimal solution in a short time with conventional optimization techniques. Hence, we propose

an actor-critic-based (AC-based) deep reinforcement learning (DRL) method to find near-optimal UAV positions at every moment. In the proposed method, the process searching for the solution iteratively at a particular moment is modeled as a Markov decision process (MDP). To handle infinite state and action spaces and improve the robustness of decision process, two powerful neural networks (NNs) are configured to evaluate

the UAV position adjustments and make decisions, respectively. Compared with heuristic, sequential least-squares programming and fixed methods, Simulation results have shown that the proposed method outperforms in terms of the throughput at every moment in UAV networks.

Massive MIMO with Binary Arrary-Receiver Based

on Constructive Noise,IEEE ICC 2020

system when employing massive binary array-receiver has been investigated while constructive noise has been observed in the single user system to detect the higher-order QAM modulated signals. To fully understand the interesting phenomenon, mathematical

model has been established and analyzed in this paper.

Theorems of the signal detectability are studied to understand the best operating signal-to-noise ratio (SNR) range based on the error behaviours of the single user SIMO system. Within the observation and analysis, a novel new multiuser SIMO with binary array-receiver structure has been proposed and can be considered as a solution to deal with the high complexity problem

that the traditional model has when using maximum likelihood (ML) detection. The key idea of this approach is to set up the multiuser multiple-input multiple-output (MIMO) model into a frequency division multiple access (FDMA) scenario and regard each user as single user SIMO to achieve the goal of decreasing

the exponentially increased complexity of ML detection method to the number of users. It is shown by numerical results that each user in this system can achieve a promising error behaviour in

the specific best operating SNR range.

Detection under Multiple Channel Models,IEEE ICC'20 Workshop - 5GLTEIC

descent (O-SGD) based learning approach is proposed to

tackle the wireless channel over-training problem inherent in artificial neural network (ANN)-assisted MIMO signal detection. Our basic idea lies in the discovery and exploitation of the training-sample orthogonality between the current training epoch and past training epochs. Unlike the conventional SGD that updates the neural network simply based upon current training samples, O-SGD discovers the correlation

between current training samples and historical training

data, and then updates the neural network with those

uncorrelated components. The network updating occurs

only in those identified null subspaces. By such means, the neural network can understand and memorize uncorrelated components between different wireless channels, and thus is more robust to wireless channel variations. This hypothesis is confirmed through our extensive computer simulations as well as performance comparison with the conventional SGD

approach.

offloading is driving the extreme utilization of available degrees of freedom (DoF) for ultra-reliable low-latency downlink communications. The fundamental aim of this work is to find latency-constrained transmission protocols that can achieve a very-low outage probability (e.g. 0:001%). Our investigation is mainly based upon the Polyanskiy-Poor-Verd´u formula on the finite-length coded channel capacity, which is extended from the

quasi-static fading channel to the frequency selective channel. Moreover, the use of a suitable duplexing mode is also critical to the downlink reliability. Specifically, time-division duplexing

(TDD) outperforms frequency-division duplexing (FDD) in terms of the frequency diversity-gain. On the other hand, FDD takes the advantage of having more temporal DoF in the downlink, which can be exchanged into the spatial diversity-gain through the use of space-time coding. Numerical study is carried out to compare the reliability between FDD and TDD under various latency constraints.